Abstract
Aims
Aging, obesity, and type 2 diabetes mellitus (T2DM) form a metabolic disease continuum that has a continuously increasing prevalence. Lipidomics explains the complex interactions between lipid metabolism and metabolic diseases. We aimed to systematically investigate the plasma lipidome changes induced by newly diagnosed impaired glucose tolerance (IGT) and T2DM in overweight/obese elderly individuals and to identify potential biomarkers to differentiate between the IGT, T2DM, and control groups.
Methods
Plasma samples from 148 overweight/obese elderly individuals, including 52 patients with IGT, 47 patients with T2DM, and 49 euglycemic controls, were analyzed using a high-coverage nontargeted absolute quantitative lipidomics approach.
Results
We quantified 1840 lipids from thirty-eight classes and seven lipid categories. Among overweight/obese elderly individuals, the lipidomic profiles of IGT and T2DM patients were significantly different from those of controls, while they were similar in the IGT and T2DM groups. The concentrations of diglycerides, triglycerides, phosphatidylcholines, and ceramides were obviously altered in the IGT and T2DM groups. Particularly, IGT and T2DM induced the accumulation of triglycerides with longer carbon atom numbers (C44–50) and saturated or lower double bond numbers (n (C=C) = 0–2). Furthermore, a total of 17 potential lipidic biomarkers were identified to successfully differentiate between the IGT, T2DM, and control groups.
Conclusions
In overweight/obese elderly patients, IGT and T2DM induced apparent lipidome-wide changes. This study’s results may contribute to explaining the complex dysfunctional lipid metabolism in aging, obesity, and diabetes.
Introduction
The growing aging population and the increasing prevalence of type 2 diabetes mellitus (T2DM) have become a substantial concern for healthcare systems worldwide. A study reported that the prevalence of diabetes peaks at the age of 65–79 years; furthermore, in 2017, there were 122 million adults aged ≥65 years with diabetes mellitus worldwide, and this number is projected to reach 253 million in 2045 (1). China has the largest number of elderly people diagnosed with diabetes (35.5 million, aged ≥60 years) (2). Impaired glucose tolerance (IGT) is a marked risk factor for T2DM, with an estimated 374 million people aged 18–99 years having IGT in 2017, and approximately half of these adults were above the age of 50 years (1). Obesity plays a critical role in the development and progression of both diabetes and prediabetes, and the body mass index (BMI) is a powerful and modifiable risk factor for T2DM (3). Hence, both aging and obesity are crucial contributing factors that lead to an imbalance between increased insulin resistance and deterioration of insulin secretory function, which results in the development and progressive worsening of T2DM (4, 5). Aging, obesity, and T2DM have formed a metabolic disease continuum with a continuously increasing prevalence (6).
Lipids perform many physiological functions, such as energy storage, cell membrane composition, and cell signaling, as well as in many diseases, such as cardiovascular diseases, neurodegenerative diseases, and cancer (7, 8, 9). Using lipidomics, researchers have identified significant biomarkers and investigated the relationship between disorders of lipid metabolism and the pathogenesis of T2DM. In a prospective study, five lipids, lysophosphatidylcholine (LPC) (18:2), triglyceride (TG) (50:1, 54:5, 56:4), and phosphatidylcholine (PC) (42:6e), were selected to predict T2DM combined with traditional risk factors (10). Even-chain saturated fatty acids (SFAs) (14:0, 16:0, and 18:0) are positively associated with incident T2D, whereas odd-chain SFAs (15:0 and 17:0) are negatively correlated with diabetes (11). Furthermore, a strong association between obesity, aging, and dysregulation of lipid metabolism has been well established. Obesity was observed to lead to increased levels of long-chain acyl-carnitines, and several lipids have been selected as biomarkers of aging, especially sphingolipids, ether-linked phospholipids, and ester-linked phospholipids (12, 13, 14). It seems complex and of great significance to explore the relationship between lipid metabolism and IGT and T2DM in obese and elderly populations.
However, complexity and lack of comprehensive coverage of the lipids are major challenges for lipidomics technologies. Ultrahigh-performance liquid chromatography (UHPLC) ranks among the most flexible and efficient separation techniques coupled with high-sensitivity detection via orbitrap mass spectrometry (UHPLC-MS/MS), allowing for the detection and identification of a broad range of lipids (15).
In the current study, we used a high-coverage nontargeted absolute quantitative lipidomics approach, UHPLC-MS/MS, to perform lipidomics in a cohort of 148 overweight/obese elderly participants, including 49 individuals with normal glucose tolerance (NGT), 52 individuals with IGT, and 47 patients with T2DM. We systematically defined the lipidomic profiles of the three groups and further investigated the IGT/T2DM-induced alteration in PCs and TGs compositions. In addition, we successfully identified potential biomarkers to discriminate between overweight/obese elderly IGT and T2DM patients and NGT subjects. The result of this study could help explain the complex dysfunctional lipid metabolism in aging, obesity, and diabetes populations.
Materials and methods
Subjects
The participants in this study were from a community-based cohort in Chengguan District, Lanzhou, China. Permanent residents aged 65–70 years with a BMI ≥24 kg/m2 were included. The exclusion criteria were as follows: use of drugs that interfered with blood glucose levels, such as glucocorticoids, metformin, and glibenclamide; use of hypolipidemic drugs within six months; current or previous history of severe hepatic disease, cancer, autoimmune diseases, or blood diseases. 99 newly diagnosed and untreated patients (including 47 patients with T2DM and 52 patients with IGT) and 49 individuals with NGT were enrolled. Each participant was asked to complete a questionnaire to collect demographic information, such as age, sex, and medical history. Informed consent was obtained from all the participants. The study was approved by the ethics committee of the Gansu Provincial Hospital (No. 2018-076).
Measurements
Clinical measurements, such as body height, weight, and blood pressure, were made by specially trained doctors and nurses using standardized methods. Plasma samples were collected from participants after an overnight fast of at least 8 h. All participants underwent a 75 g oral glucose tolerance test, and plasma glucose was obtained at 0 and 2 h during the test. T2DM was defined as FPG ≥7.0 mmol/L, or 2-h postload plasma glucose (2hPG) ≥11.1 mmol/L. IGT was defined as FPG <7.0 mmol/L, and 2hPG levels between 7.8 and 11.1 mmol/L (16).
Biochemical measurements, such as glycated hemoglobin (HbA1c), alanine transaminase (ALT), creatinine, and total cholesterol (TC) were measured using standard methods (Ci 1620 Automatic biochemical immune analysis system, Abbott Laboratories).
Lipid profiling
Plasma collection and preparation
Plasma was immediately separated from the blood samples by centrifugation for 5 min at 1000 g and stored at −80°C until analysis. Water (200 µL) was added to the thawed plasma samples at 4°C and vortexed for 5 s. Subsequently, 240 µL of precooled methanol was added to 800 μL of methyl tert-butyl ether (MTBE), vortexed again, and sonicated for 20 min. The mixture was stored at room temperature for 30 min before centrifugation at 14,000 g for 15 min. The upper layer was collected and dried under a nitrogen atmosphere. Stable isotope-labeled analogs are spiked into the samples as internal standards for quantitation. Quality control (QC) samples were prepared by mixing equal volumes of each sample.
Instrumental analysis and data preprocessing
Reverse phase chromatography (UHPLC Nexera LC-30A system) was used for sample separation using a CSH C18 column (1.7 µm, 2.1 mm × 100 mm, Waters, Taunton, Massachusetts, USA). The lipid extracts were redissolved in 200 µL 90% isopropanol/acetonitrile, centrifuged at 14,000 g for 15 min, finally, 3 µL of sample were injected. Solvent A was acetonitrile–water (6:4, v/v) with 0.1% formic acid and 0.1 mM ammonium formate and solvent B was acetonitrile–isopropanol (1:9, v/v) with 0.1% formic acid and 0.1 mM ammonium formate. The initial mobile phase was 30% solvent B at a flow rate of 300 μL/min. It was held for 2 min and then linearly increased to 100% solvent B in 23 min, followed by equilibrating at 5% solvent B for 10 min. Samples were analyzed continuously in a random order, and sampling of the queue was performed after every 10 samples by setting one of the QC samples.
Mass spectra were acquired using Q-Exactive Plus in the positive and negative modes, respectively. ESI parameters were optimized and preset for all measurements as follows: source temperature, 300°C; capillary temperature, 350°C; the ion spray voltage was set at 3000 V, S-Lens RF level was set at 50%, and the scan range of the instruments was set at m/z 200–1800. According to the methods used, which were each fully scanned (full scan), 10 pieces of map (MS2 scan, HCD) were collected, and the lipid molecules and quality of the lipid debris charge ratio were collected. The resolution of MS1 was 70,000 at m/z 200 while the resolution of MS2 was 17,500 at m/z 200.
The LipidSearch software (version 4.1, Thermo Scientific) was used for peak identification, lipid identification (metabolomics standards initiative (MSI) level 2 identification), peak extraction, peak alignment, and quantitative processing with the following parameters: precursor tolerance, 5 ppm and product ion threshold, 5%. Lipid molecules with an RSD >30% were deleted.
Statistical analysis
Clinical and laboratory measurements of the control, IGT, and T2DM groups are expressed as the mean ± standard deviation, and the proportions were analyzed using the chi-square test. The analyses were conducted using SPSS Statistics 23. Unidimensional statistical analysis was performed using the Student’s t-test, and multidimensional statistical analysis and R software map were employed for principal component analysis (PCA), orthogonal partial least squares discriminant analysis (OPLS-DA), bubble chart, hierarchical cluster analysis, and correlation analysis. The ensemble feature selection algorithm and three machine learning algorithms were applied to identify and validate the potential lipidomic biomarkers. A P < 0.05 was considered statistically significant. The data analysis was performed by R (version 4.1.0).
Results
Basic characteristics of the enrolled participants
The clinical and biochemical characteristics of the patients are summarized in Table 1. The participants aged 65–70 years, and all participants were classified as overweight (BMI ≥24 kg/m2 and BMI <28 kg/m2) or obesity (BMI ≥28 kg/m2) according to the Chinese expert consensus criteria for T2DM combined with obesity (17, 18). The levels of FPG, 2hPG, and HbA1c were significantly higher in T2DM and IGT patients than in NGT participants, while patients with T2DM exhibited higher values than those individuals with IGT.
Clinical and demographic characteristics of study participants.
| Variables | Control (n = 49) | IGT (n = 52) | T2DM (n = 47) | P |
|---|---|---|---|---|
| Females, n (%) | 25 (51.02) | 33 (63.46) | 23 (48.94) | 0.202 |
| Age (years) | 67.5 ± 1.14 | 67.3 ± 1.21 | 67.0 ± 1.26 | 0.165 |
| Current or previous smoking, n (%) | 14 (28.6) | 13 (25.0) | 16 (34.0) | 0.610 |
| Current or previous drinking, n (%) | 14 (28.6) | 11 (21.2) | 16 (34.0) | 0.354 |
| History of hypertension, n (%) | 30 (61.2) | 34 (65.4) | 33 (70.2) | 0.651 |
| Family history of diabetes, n (%) | 6 (12.2) | 17 (32.7)a | 14 (29.8)a | 0.039 |
| WC (cm) | 94.41 ± 6.56 | 94.47 ± 7.93 | 95.94 ± 8.70 | 0.441 |
| WHR | 0.91 ± 0.47 | 0.91 ± 0.58 | 0.90 ± 0.70 | 0.943 |
| WHtR | 0.58 ± 0.36 | 0.59 ± 0.52 | 0.58 ± 0.53 | 0.724 |
| BMI (kg/m2) | 26.82 ± 1.75 | 27.56 ± 2.39 | 27.28 ± 2.40 | 0.474 |
| SBP (mmHg) | 138.82 ± 16.20 | 135.71 ± 18.12 | 140.52 ± 13.44 | 0.250 |
| DBP (mmHg) | 84.82 ± 11.31 | 82.12 ± 10.55 | 83.38 ± 9.38 | 0.593 |
| FPG (mmol/L) | 5.18 ± 0.42 | 5.83 ± 0.58a | 7.05 ± 1.16a,b | < 0.001 |
| 2hPG (mmol/L) | 6.18 ± 1.03 | 9.12 ± 0.97a | 13.67 ± 3.36a,b | < 0.001 |
| HbA1c (%) | 5.21 ± 0.54 | 5.72 ± 0.52a | 7.36 ± 0.72a,b | < 0.001 |
| ALT (U/L) | 21.72 ± 9.01 | 22.21 ± 6.22 | 23.85 ± 8.29 | 0.371 |
| AST (U/L) | 23.02 ± 5.71 | 22.61 ± 6.32 | 23.95 ± 9.58 | 0.767 |
| TBil (µmol/L) | 14.56 ± 5.89 | 15.93 ± 5.17 | 16.28 ± 6.90 | 0.242 |
| Creatinine (µmol/L) | 75.09 ± 42.81 | 69.32 ± 17.16 | 66.51 ± 18.80 | 0.901 |
| BUN (mmol/L) | 5.46 ± 2.40 | 5.28 ± 1.49 | 5.05 ± 1.34 | 0.724 |
| TG (mmol/L) | 1.73 ± 0.96 | 2.07 ± 0.86 | 2.13 ± 0.70a | 0.046 |
| TC (mmol/L) | 4.99 ± 1.00 | 5.03 ± 0.99 | 5.14 ± 0.89 | 0.369 |
| HDL-C (mmol/L) | 1.33 ± 0.50 | 1.31 ± 0.47 | 1.23 ± 0.66 | 0.277 |
| LDL-C (mmol/L) | 2.63 ± 0.75 | 2.62 ± 0.81 | 2.72 ± 0.15 | 0.857 |
Data presented as mean (s.d.) for continuous variables or number (%) for categorical variables.
aCompared to control group, P < 0.05; bCompared to IGT group, P < 0.05.
2hPG, 2-h postload plasma glucose; ALT, alanine transaminase; AST, aspartate transaminase; BMI, body mass index; BUN, urea nitrogen; DBP, diastolic blood pressure; FPG, fasting plasma glucose; HbA1c, glycated hemoglobin; HDL-C, high-density lipoprotein cholesterol; IGT, impaired glucose tolerance; LDL-C, low-density lipoprotein cholesterol; SBP, systolic blood pressure; TBil, total bilirubin; T2DM, type 2 diabetes mellitus; TC, total cholesterol; TG, total triglyceride; WC, waist circumference; WHR, waist-to-hip ratio; WHtR, waist-to-height ratio.
Plasma lipidomic profiles of NGT, IGT, and T2DM in overweight/obese elderly individuals
In this high-coverage nontargeted absolute quantitative lipidomics analysis, 1840 lipids (1274 lipids in positive modes and 566 lipids in negative modes) from 38 classes and seven lipid categories were detected (Fig. 1A). We compared concentrations of some of the main lipid classes in plasma from IGT and T2DM patients to those from euglycemic controls using a t-test analysis (Fig. 1B). The concentrations of dihexosyl N-acetylhexosyl ceramide (CerG3GNAc1) and trihexosyl di-N-acetylhexosyl ceramide (CerG3GNAc2) classes were significantly decreased, while those of diglyceride (DG), phosphatidic acid (PA), PC, and phosphatidylethanolamine (PE) classes were significantly increased in subjects with ITG and T2DM. Lipids with a fold change (FC) of >1.5 or <0.67 and a false discovery rate (FDR) of <0.01, were selected, and the differential lipids were displayed using a Venn diagram (Fig. 1C, Supplementary Tables 1, 2, and 3, see section on supplementary materials given at the end of this article). In the IGT, T2DM, and control groups, 19 lipids were found to be the common differential lipids of the three groups, and most of them were glycerophospholipids. Moreover, 78% of the common differential lipids induced by IGT and T2DM were glycerolipids (DGs and TGs). To further investigate whether clinical parameters are associated with lipidomic alterations identified by the lipidomics, we calculated the Spearman correlations between the lipid classes and the relevant clinical parameters (FPG, 2hPG, HbA1c, BMI, WC, WHR, WHtR, age, and gender). The heat map (Fig. 1D) showed that levels of TG and DG were positively correlated with nearly all the relevant clinical parameters but negatively correlated with female. On the other hand, PI, PIP, LPI with a positive correlation with glucose related parameters (FPG, 2hPG, and HbA1c), but with inverse correlation with obesity-related parameters (BMI, WC, WHR, and WHtR). Sphingolipids (including Cer, CerP, SM, phSM, GM1, and GM2) tended to be negatively correlated with obesity and glucose related parameters.
(A) Bar plots for numbers of quantified lipids in overweight/obese elderly IGT, T2DM, and euglycemic controls plasma samples. (B) Heat map representation of the foldchanges in the concentration of every class of lipids in plasma from IGT patients compared to controls. (*P < 0.05, **P < 0.01. t-test). (C) Venn diagram depicting the number of significantly changed lipids from three paired comparisons (fold change (FC) >1.5 or <0.67 and false discovery rate (FDR) <0.05). In addition, the common differential lipids from the two paired comparisons of IGT vs normal and T2DM vs normal were also shown. (D) Heat map representation of the Spearman’s correlations between the levels of lipid classes and the relevant clinical parameters (FPG, 2hPG, HbA1c, BMI, WC, WHR, WHtR, age, and gender).
Citation: Endocrine Connections 12, 12; 10.1530/EC-23-0212
Lipid alterations in the plasma of overweight/obese elderly patients with IGT
OPLS-DA was performed to achieve maximum separation between groups and mine the differential lipid molecules with biological significance. In OPLS-DA, the two groups showed obvious classified aggregation on the scatter plot without overfitting, indicating that there was a significant difference in lipidomic patterns between the IGT and control groups (Fig. 2A, Supplementary Fig. 1A). A total of 423 lipids with significant differences were screened according to a P-value of <0.05 and variable importance in the projection (VIP) >1, which are generally considered to contribute significantly to model interpretation.
Lipidome profiling changes of IGT. (A) Scores scatter plot for OPLS-DA of plasma lipidomic profiling in IGT and Normal group. (B) The relative percentage difference in concentration of main lipid species between IGT and normal group. Each dot represents a lipid species, and the different lipid classes are color-coded. Bubble size indicates significance. Smaller bubbles indicate significant differences (0.01 < P < 0.05), larger bubbles indicate highly significant differences (P < 0.01). (C) Hierarchical clustering of the lipid–lipid correlation matrix and lipid species in clusters B and C, %. Rows and columns correspond to the 423 measured lipid species. Total concentration of phosphatidylcholine (PC) (D) and triglyceride (TG) class (G). Distribution of PC and TG class in normal and IGT group classified according to the number of carbon atoms (E, H) and degree of saturation (F, I). Values represent plasma concentrations of lipid on a logarithmic scale. Indicated are P-values for comparisons for each species (t-test). *P <0.05, **P < 0.01. Error bars denote means ± s.d.
Citation: Endocrine Connections 12, 12; 10.1530/EC-23-0212
Figure 2B shows significant IGT-induced changes in the abundance of lipid species in most of the analyzed lipid classes. Compared to the control group, the levels of most lipid species (especially lipids belonging to TG, PC, sphingomyelin (SM), and DG classes) in the IGT group increased, while the levels of lipids belonging to ceramide (Cer) and CerG3GNAc1 classes decreased. Correlation analysis could be useful in measuring the metabolic proximities of significantly changed lipids, which would help to further the current understanding of the interregulatory relationship between lipids in the process of biological state change (19). Hierarchical clustering of the lipid–lipid correlation matrix based on the significantly different lipids was performed. Four distinct lipid clusters have been identified. The lipids within clusters demonstrated diverse compositions, suggesting that clustered lipids might be coregulated or functionally related, and the percentages of lipid class in the two obvious clusters (B and C) were shown in Fig. 2C.
We assessed the IGT-related PC and TG compositions by analyzing the saturation and chain length of the PC and TG species (Fig. 2D, E, F, G, H and I). IGT induced elevated total concentrations of PCs, and further analysis showed that the saturation and chain length of PC-associated fatty acids changed significantly. The concentrations of PCs with shorter carbon atom numbers (C17-30) and longer even chains (C32, C34, C36, C38, C40, C42, C50, C52, and C54) were elevated in the IGT group (Fig. 2E). The total TG content was similar in the control and IGT groups, but significantly increased levels of TGs containing longer chain fatty acids were observed in the IGT group (Fig. 2H); furthermore, patients in the IGT group had significantly more TGs with two or fewer double bonds (Fig. 2I).
Lipid alterations in the plasma of overweight/obese elderly patients with T2DM
The OPLS-DA analysis separated the T2DM and control groups with acceptable values (R2Y:0.9323) (Fig. 3A and Supplementary Fig. 1B). Using this approach, 535 lipids were identified, with significant differences between the two groups. The bubble chart (Fig. 3B) revealed T2DM-induced obvious changes in lipid species, especially lipids belonging to the TG, SM, phosphatidylserine (PS), PC, DG, and Cer classes. Consistent with the observed changes in patients with IGT, patients with T2DM had reduced plasma levels of lipids in the Cer and CerG3GNAc1 classes. Hierarchical clustering of the lipid–lipid correlation matrix indicated strong positive correlations within 535 significantly different lipid species. Five distinct lipid clusters were identified, and the lipids within clusters B and C were also identified (Fig. 3C). The overall abundance of TGs and PCs, which are the most and second most abundant lipid classes in the plasma, respectively, were both affected by T2DM (Fig. 3D and G). T2DM, which was similar to IGT, induced increased concentrations of longer even-chain (C32, C34, C36, C38, C40, and C52) PCs and elevated levels of longer length (C44-52, C56, and C58) TGs (Fig. 3E and H). Surprisingly, the levels of PCs with shorter chain lengths (C21, C24, C27, C29, and C30) decreased in the T2DM group, which was not consistent with those of the IGT group. We found that saturation profiles differed in TGs, with patients with T2DM having a significant increase in TGs with two or fewer double bonds (Fig. 3I).
Lipidome profiling changes of T2DM. (A) Scores scatter plot for OPLS-DA of plasma lipidomic profiling in T2DM and Normal group. (B) The relative percentage difference in concentration of main lipid species between T2DM and Normal group. Each dot represents a lipid species, and the different lipid classes are color coded. Bubble size indicates significance. Smaller bubbles indicate significant differences (0.01 < P < 0.05), larger bubbles indicate highly significant differences (P < 0.01). (C) Hierarchical clustering of the lipid–lipid correlation matrix and lipid species in clusters B and C. Rows and columns correspond to the 535 measured lipid species. Total concentration of phosphatidylcholine (PC) (D) and triglyceride (TG) class (G). Distribution of PC and TG class in normal and T2DM group classified according to the number of carbon atoms (E, H) and degree of saturation (F, I). Values represent plasma concentrations of lipid on a logarithmic scale. Indicated are P-values for comparisons for each species (t-test). *P < 0.05, **P < 0.01. Error bars denote means ± s.d.
Citation: Endocrine Connections 12, 12; 10.1530/EC-23-0212
Comparison of the plasma lipidomic profiles of the IGT and T2DM groups
The OPLS-DA analysis showed a separation of lipid metabolites in the IGT and T2DM groups without overfitting (Fig. 4A and Supplementary Fig. 1C). Among the 255 lipid species with VIP >1 and P < 0.05, 17%, 10%, and 18% were TGs, PCs, and Cers, respectively. Figure 4B and C show the lipid class distributions of these significantly different lipid species and lipid–lipid correlation matrix. The overall abundance of CerG3GNAc1 in patients with IGT was higher than those in patients with T2DM, and the saturation and chain length of CerG3GNAc1s changed significantly (Fig. 4D, E and F).
Lipidome profiling changes between IGT and T2DM. (A) Scores scatter plot for OPLS-DA of plasma lipidomic profiling in IGT and T2DM group. (B) The relative percentage difference in concentration of main lipid species between IGT and T2DM group. Each dot represents a lipid species, and the different lipid classes are color coded. Bubble size indicates significance. Smaller bubbles indicate significant differences (0.01 < P < 0.05), larger bubbles indicate highly significant differences (P < 0.01). (C) Hierarchical clustering of the lipid-lipid correlation matrix and lipid species in clusters C and F. Rows and columns correspond to the 255 measured lipid species. (D) Total concentration of dihexosyl N-acetylhexosyl ceramide (CerG3GNAc1) class. Distribution of CerG3GNAc1 class in IGT and T2DM group classified according to the number of carbon atoms (E) and degree of saturation (F). Values represent plasma concentrations of lipid on a logarithmic scale. Indicated are P-values for comparisons for each species (t-test). *P < 0.05, **P < 0.01. Error bars denote means ± s.d.
Citation: Endocrine Connections 12, 12; 10.1530/EC-23-0212
Identification of lipidomic biomarkers for IGT and T2DM in overweight/obese elderly individuals
After systematically defining the lipidomic profiles associated with IGT and T2DM, we developed prediction models by selecting lipidomic biomarkers that can be used to differentiate individuals with IGT and patients with T2DM from euglycemic controls. Here an ensemble learning-based feature selection technique that integrates statistical tests and multiple feature selection methods (Filter, Wrapper, and Embedded) were used. Therefore, a group of robust diagnostic lipidomic small molecules can be selected to build a more effective diagnostic marker panel model. In brief, reward those lipidomic small molecules that were frequently selected in the feature selection method, and the comprehensive weight values of each lipidomic small molecule were calculated by combining the reward scores obtained in each feature selection method. The larger the weight values indicated the greater the contribution of the lipidomic small molecules in distinguishing the experimental group and control samples. We identified the 10 most important differential lipidomic small molecules (LPE (20:4), Cer (m22:0/18:0), PA (18:2/10:4), PS (14:0/14:0), CerG3GNAc1 (d34:1), CerG3GNAc1 (t37:6), Cer (m32:0), TG (20:4e/12:2/12:2), Cer (m35:0), ChE (20:5)) between control group and IGT group, the top three important differential lipidomic small molecules (PA (18:2/10:4), Cer (m32:0), and PG (30:3)) between control and T2DM groups, and the four most important differential lipidomic small molecules (Cer (m32:0), CerG3GNAc1 (d34:1), CerG3GNAc1 (t37:6), and PC (10:0/11:4)) between the IGT and T2DM groups. The importance of the selected biomarkers in random forest (RF) is shown in Supplementary Fig. 2A, B, and C. In addition, receiver operating characteristic (ROC) analysis was performed to evaluate the effect of each substance on the area under the curve (AUC) of the models. The AUC demonstrated that the selected biomarkers contributed significantly to the classification ability (Supplementary Fig. 2D, E, and F), and the box plots show the abundance of these lipidomic biomarkers in the prediction models (Fig. 5A, C, and E).
Differential expression levels of 17 potential biomarkers and receiver operator characteristic (ROC) curves for model evaluations. (A, C, E) The expression levels of potential biomarkers in different groups. (B, D, F) Receiver operating characteristic area under the curves (AUCs) for sensitivity and specificity of the predictive models for discriminating IGT and normal, T2DM and normal, and IGT and T2DM respectively. LR, logistic regression; RF, random forest; SVM, support vector machine.
Citation: Endocrine Connections 12, 12; 10.1530/EC-23-0212
Finally, three commonly used machine learning algorithms, logistic regression (LR), RF, and support vector machine (SVM), were used to verify the screening results. LR, RF, and SVM yielded good AUCs (0.84–0.91) in differentiating patients with IGT from controls, although RF appeared slightly better (Fig. 5B). Potential biomarkers, PA (18:2/10:4), Cer (m32:0), and PG (30:3) achieved nearly excellent AUC values from 0.89 to 0.98 (Fig. 5D). Additionally, the prediction model differentiating IGT from T2DM yielded good AUCs of 0.84–0.86 using the three machine learning algorithms (Fig. 5F).
Discussion
Aging and obesity are crucial contributing factors to the development and progressive worsening of T2DM. The large population of obese elderly people with diabetes has caused a tremendous health crisis worldwide. Using lipidomics, many lipid metabolites have been detected and implicated as critical components that can be used to explain the complex interactions between aging, obesity, insulin resistance (IR), and T2DM (6).
To identify associations between lipids and T2DM incidence, several large-scale lipidomic cohort studies have been performed in adults (20, 21, 22, 23). Given the large number of obese older adults with T2DM, our study included only a ‘pure’ population of patients with obesity adults aged 65–70 years with NGT, IGT, and T2DM and presented a comprehensive lipidomic evaluation. We aimed to explore the lipidomic patterns of IGT and T2DM in obesity elderly population. At the same time, we delved into the correlation between the lipid species and the relevant clinical parameters, such as BMI, age, FPG, and 2hPG. These results may contribute to explaining the complex dysfunctional lipid metabolism in aging, obesity, and diabetes among obese elderly T2DM patients.
We found that patients with IGT and T2D had increased concentrations of DG, TG, PC, sphingomyelins (SM), and Cer species. These results are in agreement with the findings of cross-sectional studies and prospective studies, which revealed that IGT and T2D were positively associated with dysglycemia and T2DM (10, 20, 21, 24).
TGs and DGs showed the most significant association with T2D and IGT. Our analysis is in line with those of previous studies, which showed that TG(46:1), TG(48:1), TG(48:2), TG(51:1), and TG (52:1) were positively associated with T2DM and prediabetes in an American Indian cohort (22), and positively associated with T2DM in Chinese (21) and Caucasian cohorts (25). In addition, previous studies (14, 26) also assessed the effects of age and obesity on the serum levels of TG and DG species, which supported our findings that levels of TG and DG were positively correlated with obesity- and glucose-related parameters.
Chain length and desaturation of fatty acid moieties in lipid molecules complicate the assignment of biological roles to lipid classes (27). In the PREDIMED trial (28), odd-chain TGs (C53, C55) were negatively associated with diabetes risk while TGs with 48–50 carbon atom numbers and two to three double bonds as risk factors for the development of T2DM (21). Additionally, a cluster of TG species with saturated and monounsaturated acyl chains were identified to be associated with the prevalence and incidence of DM (25, 29). Increased levels of long-chain SFAs are known to contribute to insulin resistance and T2DM (30). We systematically examined IGT- and T2DM-associated alterations in the number of carbon atoms and double bonds in the various lipid classes that were investigated. Significant changes in the number of carbon atoms and the degree of unsaturation were observed in the PCs and TGs. Patients with IGT and T2DM tend to have increased levels of TGs with longer carbon atom numbers (C44–50), saturated or lower double bond numbers (n (C=C) = 0–2).
The lesser metabolic and oxidation rates of monounsaturated and long-chain SFAs affecting body weight and composition in obesity T2DM patients were the possible mechanisms (31). TG- and DG-mediated insulin resistance is the unifying molecular mechanism that explains the most common forms of IR associated with obesity and aging, as well as T2DM (26, 27, 32). Different views were put forward when researchers discovered that inhibiting PE production and the subsequent accumulation of DG and TG retained insulin sensitivity and increased mitochondrial biogenesis and muscle oxidative capacity in knockout mouse muscles (33). They believe that phospholipids, rather than DGs or TGs, are probable modulators of IR in muscles (6).
We also found that the levels of multiple glycerophospholipids, including PC (10:0/11:4), PC (14:0/10:1), PS (11:0/16:0), PS (11:0/18:0), phosphatidylinositol (PI) (16:0/16:0), and PI (16:0/16:1), changed significantly in individuals with IGT and T2DM. Although previous studies have reported associations of PC, PE, PS, and PI species with IR, T2D, and related traits (21, 22, 29, 34), some studies have shown inconsistent results as PCs and PEs were increased in some studies (35, 36) but were decreased in others (24, 37). In young adults (aged 18–34 years), independent of age and BMI, PI (16:0/16:0) and PI (16:0/16:1) were positively associated with insulin AUC in men and homeostatic model assessment of insulin resistance (HOMA-IR) in both women and men (34). PI (16:0/16:1) was also found to be positively associated with prediabetes over a 5-year follow-up period (22).
Furthermore, for PCs, we observed that IGT and T2DM induced increased concentrations of even-chain (C32, C34, C36, C38, C40, and C52) PCs. In contrast to the IGT group, the levels of PCs with shorter chain lengths (C21, C24, C27, C29, and C30) decreased in the T2DM group. A relationship between diabetes risk and the carbon number and double bond content among PCs was also identified by Rhee et al. They depicted a downsloping pattern in which PCs with relatively lower carbon numbers and double bond content were most significantly elevated in patients with T2DM compared to the levels in controls (25). LPCs, such as LPC (18:0), LPC (18:1), and LPC (18:2), were negatively correlated with T2DM, while no significant change in LPCs was found in the present study.
Cer is the precursor of ganglioside and SM, and lipidomic profiling has revealed relationships between their levels, aging, obesity, and diabetes (6, 14). In the present study, we found that the levels of most Cers in the IGT group were reduced and the levels of gangliosides GM3 and SMs were elevated (Fig. 2B). Compared with IGT, T2DM induced a higher proportion of Cers, less increased SMs, and slightly decreased ganglioside GM3 (Fig. 3B). These results suggest a shift in the balance of sphingolipid metabolism as diabetes progresses. Further, saturated SMs (C34:0, C36:0, C38:0, C40:0) and unsaturated sphingomyelins (C34:1, C36:1, C42:3) were reported to be risk factors for IR and incident T2D among 1974 ethnically Chinese individuals (38). Our study showed that apart from saturated SM(d36:0), SM(d38:0) and SM(d39:0), the levels of SM(d42:7) and SM(d44:4) were also higher in patients with T2DM but not in patients with IGT.
Machine learning algorithms have been increasingly recognized as enabling techniques for selecting biomarkers for various human diseases (39). Finally, we successfully selected potential biomarkers to distinguish the NGT, IGT, and T2DM groups in overweight/obese elderly individuals using three machine learning algorithms. The prediction models differentiate the three groups yielding good AUCs. Interestingly, PA (18:2/10:4) and Cer (m32:0) were the common biomarkers of IGT and T2DM. The levels of PA (18:2/10:4) were elevated in IGT patients but reduced in subjects with T2DM, while the concentration of Cer (m32:0) showed a stepwise upward trend in IGT and T2DM groups. These findings suggest that the lipidomic changes at different stages of the diabetes are certainly far more complex than we expected. Therefore, future studies are warranted to investigate the roles of lipids in diabetes, aging, and obesity in longitudinal cohorts.
The current study has several limitations. First, although a high-coverage and the most flexible and efficient system we used, we carried out only nontargeted lipidomics analysis. The sample size of this study was also relatively small. Second, to ensure the stability and reliability of the lipidomics analysis, we carried out a strict quality control evaluation of the results, while the cross-sectional study that lacked prospective lipidomics data weakened the credibility of the results.
In summary, our high-coverage nontargeted absolute quantitative lipidomic analysis revealed novel lipidomic patterns in overweight/obese elderly individuals with IGT and T2DM. Particularly, IGT and T2DM induced the accumulation of triglycerides with longer carbon atom numbers (C44–50) and saturated or lower double bond numbers (n (C=C) = 0–2). A panel of differential lipids was successfully identified as a potential biomarker in patients with IGT and T2DM. The lipidomic profile may improve our understanding of the extent and complexity of lipid dysregulation in obesity, aging, and diabetes and provide new insights into the underlying molecular mechanisms of diabetes.
Supplementary materials
This is linked to the online version of the paper at https://doi.org/10.1530/EC-23-0212.
Declaration of interest
There is no conflict of interest that could be perceived as prejudicing the impartiality of the study reported.
Funding
This work was supported by Natural Science Foundation of Gansu province, China (No. 22ZD6FA033 and No. 22JR5RA672) and Science and Technology Foundation for Young Scientists of Gansu province, China (No. 21JR7RA652).
Ethical approval and consent by participants
This study was approved by the Ethics Committee of Gansu Provincial Hospital (No. 2018-076). All participants provided written informed consent prior to their inclusion in the study.
Data availability
The datasets used and/or analyzed during the current study are not publicly available due to the individual privacy of the patients included in this study but are available from the corresponding author on reasonable request.
Author contribution statement
Limin Tian conceived the study and reviewed/edited the manuscript. Feifei Shao analyzed the data and wrote the manuscript. Xinxin Hu, Jiayu Li, and Bona Bai contributed to the collection of samples and discussion on the manuscript. All authors contributed to the article and approved the submitted version of the manuscript.
Acknowledgements
We are grateful to all participants for their dedication to data collection and laboratory measurements.
References
- 1↑
Cho NH, Shaw JE, Karuranga S, Huang Y, da Rocha Fernandes JD, Ohlrogge AW, & Malanda B. IDF Diabetes Atlas: global estimates of diabetes prevalence for 2017 and projections for 2045. Diabetes Research and Clinical Practice 2018 138 271–281. (https://doi.org/10.1016/j.diabres.2018.02.023)
- 2↑
International Diabetes Federation (IDF). IDF Diabetes Atlas, 8th ed. 2017. Brussels, Belgium: International Diabetes Federation (IDF), 2017. (available at: https://diabetesatlas.org/upload/resources/previous/files/8/IDF_DA_8e-EN-final.pdf)
- 3↑
Neeland IJ, Turer AT, Ayers CR, Powell-Wiley TM, Vega GL, Farzaneh-Far R, Grundy SM, Khera A, McGuire DK, & de Lemos JA. Dysfunctional adiposity and the risk of prediabetes and type 2 diabetes in obese adults. JAMA 2012 308 1150–1159. (https://doi.org/10.1001/2012.jama.11132)
- 4↑
Mancuso P, & Bouchard B. The impact of aging on adipose function and adipokine synthesis. Frontiers in Endocrinology (Lausanne) 2019 10 137. (https://doi.org/10.3389/fendo.2019.00137)
- 5↑
Martyniak K, & Masternak MM. Changes in adipose tissue cellular composition during obesity and aging as a cause of metabolic dysregulation. Experimental Gerontology 2017 94 59–63. (https://doi.org/10.1016/j.exger.2016.12.007)
- 6↑
Meikle PJ, & Summers SA. Sphingolipids and phospholipids in insulin resistance and related metabolic disorders. Nature Reviews. Endocrinology 2017 13 79–91. (https://doi.org/10.1038/nrendo.2016.169)
- 7↑
Xiang Y, Lam SM, & Shui G. What can lipidomics tell us about the pathogenesis of Alzheimer disease? Biological Chemistry 2015 396 1281–1291. (https://doi.org/10.1515/hsz-2015-0207)
- 8↑
Pechlaner R, Kiechl S, & Mayr M. Potential and caveats of lipidomics for cardiovascular disease. Circulation 2016 134 1651–1654. (https://doi.org/10.1161/CIRCULATIONAHA.116.025092)
- 9↑
Yu GP, Chen GQ, Wu S, Shen K, & Ji Y. The expression of PEBP4 protein in lung squamous cell carcinoma. Tumour Biology 2011 32 1257–1263. (https://doi.org/10.1007/s13277-011-0230-1)
- 10↑
Suvitaival T, Bondia-Pons I, Yetukuri L, Pöhö P, Nolan JJ, Hyötyläinen T, Kuusisto J, & Orešič M. Lipidome as a predictive tool in progression to type 2 diabetes in Finnish men. Metabolism: Clinical and Experimental 2018 78 1–12. (https://doi.org/10.1016/j.metabol.2017.08.014)
- 11↑
Forouhi NG, Koulman A, Sharp SJ, Imamura F, Kröger J, Schulze MB, Crowe FL, Huerta JM, Guevara M, Beulens JW, et al.Differences in the prospective association between individual plasma phospholipid saturated fatty acids and incident type 2 diabetes: the EPIC-InterAct case-cohort study. Lancet. Diabetes and Endocrinology 2014 2 810–818. (https://doi.org/10.1016/S2213-8587(1470146-9)
- 12↑
Mihalik SJ, Goodpaster BH, Kelley DE, Chace DH, Vockley J, Toledo FG, & DeLany JP. Increased levels of plasma acylcarnitines in obesity and type 2 diabetes and identification of a marker of glucolipotoxicity. Obesity 2010 18 1695–1700. (https://doi.org/10.1038/oby.2009.510)
- 13↑
Yang S, Dong Y, Liu Y, Yan X, Sun G, Jia G, Li X, Liu H, Su H, & Li Y. Application of lipidomics strategy to explore aging-related biomarkers and potential anti-aging mechanisms of ginseng. Biogerontology 2021 22 589–602. (https://doi.org/10.1007/s10522-021-09937-8)
- 14↑
Kawanishi N, Kato Y, Yokozeki K, Sawada S, Sakurai R, Fujiwara Y, Shinkai S, Goda N, & Suzuki K. Effects of aging on serum levels of lipid molecular species as determined by lipidomics analysis in Japanese men and women. Lipids in Health and Disease 2018 17 135. (https://doi.org/10.1186/s12944-018-0785-6)
- 15↑
Züllig T, & Köfeler HC. High resolution mass spectrometry in lipidomics. Mass Spectrometry Reviews 2021 40 162–176. (https://doi.org/10.1002/mas.21627)
- 16↑
Alberti KG, & Zimmet PZ. Definition, diagnosis and classification of diabetes mellitus and its complications. Part 1: diagnosis and classification of diabetes mellitus provisional report of a WHO consultation. Diabetic Medicine 1998 15 539–553. (https://doi.org/10.1002/(SICI)1096-9136(199807)15:7<539::AID-DIA668>3.0.CO;2-S)
- 17↑
Endocrinology Branch of the Chinese Medical Association. Expert consensus on integrated management of type 2 diabetes mellitus complicated with obesity in China. Chinese Journal of Diabetes Mellitus 2016 8 662–666. (https://doi.org/10.3760/cma.j.issn.1674-5809.2016.11.006)
- 18↑
Chinese Medical Association, Chinese Medical Journals Publishing House, Chinese Society of General Practice, Editorial Board of Chinese Journal of General Practitioners of Chinese Medical Association, Expert Group of Guidelines for Primary Care of Endocrine System Disease, Hong L & Shanzhu Z. Guideline for primary care of obesity (2019). Chinese Journal of General Practitioners 2020 19 95–101. (https://doi.org/10.3760/cma.j.issn.1671-7368.2020.02.002)
- 19↑
Koberlin MS, Snijder B, Heinz LX, Baumann CL, Fauster A, Vladimer GI, Gavin AC, & Superti-Furga G. A conserved circular network of coregulated lipids modulates innate immune responses. Cell 2015 162 170–183. (https://doi.org/10.1016/j.cell.2015.05.051)
- 20↑
Chew WS, Torta F, Ji S, Choi H, Begum H, Sim X, Khoo CM, Khoo EYH, Ong WY, Van Dam RM, et al.Large-scale lipidomics identifies associations between plasma sphingolipids and T2DM incidence. JCI Insight 2019 5 e126925. (https://doi.org/10.1172/jci.insight.126925)
- 21↑
Lu J, Lam SM, Wan Q, Shi L, Huo Y, Chen L, Tang X, Li B, Wu X, Peng K, et al.High-coverage targeted lipidomics reveals novel serum lipid predictors and lipid pathway dysregulation antecedent to Type 2 diabetes onset in normoglycemic Chinese adults. Diabetes Care 2019 42 2117–2126. (https://doi.org/10.2337/dc19-0100)
- 22↑
Miao G, Zhang Y, Huo Z, Zeng W, Zhu J, Umans JG, Wohlgemuth G, Pedrosa D, DeFelice B, Cole SA, et al.Longitudinal plasma lipidome and risk of Type 2 diabetes in a large sample of American Indians with normal fasting glucose: the strong heart family study. Diabetes Care 2021 44 2664–2672. (https://doi.org/10.2337/dc21-0451)
- 23↑
Fernandez C, Surma MA, Klose C, Gerl MJ, Ottosson F, Ericson U, Oskolkov N, Ohro-Melander M, Simons K, & Melander O. Plasma lipidome and prediction of Type 2 diabetes in the population-based Malmo diet and cancer cohort. Diabetes Care 2020 43 366–373. (https://doi.org/10.2337/dc19-1199)
- 24↑
Floegel A, Stefan N, Yu Z, Mühlenbruch K, Drogan D, Joost HG, Fritsche A, Häring HU, Hrabě de Angelis M, Peters A, et al.Identification of serum metabolites associated with risk of type 2 diabetes using a targeted metabolomic approach. Diabetes 2013 62 639–648. (https://doi.org/10.2337/db12-0495)
- 25↑
Rhee EP, Cheng S, Larson MG, Walford GA, Lewis GD, McCabe E, Yang E, Farrell L, Fox CS, O'Donnell CJ, et al.Lipid profiling identifies a triacylglycerol signature of insulin resistance and improves diabetes prediction in humans. Journal of Clinical Investigation 2011 121 1402–1411. (https://doi.org/10.1172/JCI44442)
- 26↑
Erion DM, & Shulman GI. Diacylglycerol-mediated insulin resistance. Nature Medicine 2010 16 400–402. (https://doi.org/10.1038/nm0410-400)
- 27↑
Yang Q, Vijayakumar A, & Kahn BB. Metabolites as regulators of insulin sensitivity and metabolism. Nature Reviews. Molecular Cell Biology 2018 19 654–672. (https://doi.org/10.1038/s41580-018-0044-8)
- 28↑
Razquin C, Toledo E, Clish CB, Ruiz-Canela M, Dennis C, Corella D, Papandreou C, Ros E, Estruch R, Guasch-Ferre M, et al.Plasma lipidomic profiling and risk of Type 2 diabetes in the PREDIMED trial. Diabetes Care 2018 41 2617–2624. (https://doi.org/10.2337/dc18-0840)
- 29↑
Meikle PJ, Wong G, Barlow CK, Weir JM, Greeve MA, MacIntosh GL, Almasy L, Comuzzie AG, Mahaney MC, Kowalczyk A, et al.Plasma lipid profiling shows similar associations with prediabetes and type 2 diabetes. PLoS One 2013 8 e74341. (https://doi.org/10.1371/journal.pone.0074341)
- 30↑
Ariyama H, Kono N, Matsuda S, Inoue T, & Arai H. Decrease in membrane phospholipid unsaturation induces unfolded protein response. Journal of Biological Chemistry 2010 285 22027–22035. (https://doi.org/10.1074/jbc.M110.126870)
- 31↑
Im DS. FFA4 (GPR120) as a fatty acid sensor involved in appetite control, insulin sensitivity and inflammation regulation. Molecular Aspects of Medicine 2018 64 92–108. (https://doi.org/10.1016/j.mam.2017.09.001)
- 32↑
Castoldi A, Monteiro LB, van Teijlingen Bakker N, Sanin DE, Rana N, Corrado M, Cameron AM, Hassler F, Matsushita M, Caputa G, et al.Triacylglycerol synthesis enhances macrophage inflammatory function. Nature Communications 2020 11 4107. (https://doi.org/10.1038/s41467-020-17881-3)
- 33↑
Selathurai A, Kowalski GM, Burch ML, Sepulveda P, Risis S, Lee-Young RS, Lamon S, Meikle PJ, Genders AJ, McGee SL, et al.The CDP-ethanolamine pathway regulates skeletal muscle diacylglycerol content and mitochondrial biogenesis without altering insulin sensitivity. Cell Metabolism 2015 21 718–730. (https://doi.org/10.1016/j.cmet.2015.04.001)
- 34↑
Beyene HB, Hamley S, Giles C, Huynh K, Smith A, Cinel M, Mellet NA, Morales-Scholz MG, Kloosterman D, Howlett KF, et al.Mapping the associations of the plasma lipidome with insulin resistance and response to an oral glucose tolerance test. Journal of Clinical Endocrinology and Metabolism 2020 105 e1041–e1055. (https://doi.org/10.1210/clinem/dgaa054)
- 35↑
Huang Q, Yin P, Wang J, Chen J, Kong H, Lu X, & Xu G. Method for liver tissue metabolic profiling study and its application in type 2 diabetic rats based on ultra performance liquid chromatography-mass spectrometry. Journal of Chromatography. B, Analytical Technologies in the Biomedical and Life Sciences 2011 879 961–967. (https://doi.org/10.1016/j.jchromb.2011.03.009)
- 36↑
Drogan D, Dunn WB, Lin W, Buijsse B, Schulze MB, Langenberg C, Brown M, Floegel A, Dietrich S, Rolandsson O, et al.Untargeted metabolic profiling identifies altered serum metabolites of type 2 diabetes mellitus in a prospective, nested case control study. Clinical Chemistry 2015 61 487–497. (https://doi.org/10.1373/clinchem.2014.228965)
- 37↑
Xu F, Tavintharan S, Sum CF, Woon K, Lim SC, & Ong CN. Metabolic signature shift in type 2 diabetes mellitus revealed by mass spectrometry-based metabolomics. Journal of Clinical Endocrinology and Metabolism 2013 98 E1060–E1065. (https://doi.org/10.1210/jc.2012-4132)
- 38↑
Yu B, Yun H, Sun L, Wu Q, Zong G, Qi Q, Li H, Zheng H, Zeng R, Liang L, et al.Associations among circulating sphingolipids, β-cell function, and risk of developing type 2 diabetes: a population-based cohort study in China. PLOS Medicine 2020 17 e1003451. (https://doi.org/10.1371/journal.pmed.1003451)
- 39↑
Shen X, Wang C, Liang N, Liu Z, Li X, Zhu ZJ, Merriman TR, Dalbeth N, Terkeltaub R, Li C, et al.Serum metabolomics identifies dysregulated pathways and potential metabolic biomarkers for hyperuricemia and gout. Arthritis and Rheumatology 2021 73 1738–1748. (https://doi.org/10.1002/art.41733)
